2P1-E19 ロボットビジョンによる室内環境の認識 : SIFT特徴量とエッジ情報を用いたオフィス機器の画像探索と認識

Abstract

For the purpose of simultaneous localization and mapping (SLAM) of an indoor mobile robot, a method to process only one monocular image (640×480) of the whole view of environment is proposed. This method imitates that a human can grasp momentarily the situation of a room, i.e., the layout of objects/obstacles. Specific object recognition for desks with several view angles are dealt with. The proposed method is concretely organized as follows: 1) Bag of Keypoints is applied, then the existence of object on input image is judged, 2) If object existence is verified, its angle is detected by Bag of Keypoints, 3) The candidates for the projection from template image to input image is obtained using Scale Invariant Feature Transform (SIFT) and edge information, the validity of which is decided by AdaBoost classifier based on various image features, such as Hu moment, Haar-like features, etc.